Misleading claims identified from evidence
100% (95% CI 86–100%, n=24)
MEASURED TEST RESULTS
How the checker performed.
A small, repeatable test of cited claims against public-source facts. Every case and source is listed below.
What this measures: 36 human-authored, pre-labeled single-draft cases against four short, paraphrased snapshots of official U.S. agency pages. The full extraction, cited-passage verification, and deterministic gate run on each case. This is diagnostic, not a production-accuracy estimate; labels have not been independently adjudicated.
What changed: On the first run, 23/36 cases were resolved and scored correctly; 13 cases lost a citation when a sentence was split into atomic claims. After the citation-mapping fix, this run has 36/36 correct with 0 unresolved citations. The fixture, prompt, model, and gate versions were unchanged.
Correct claims accepted
100% (95% CI 76–100%, n=12)
Runs with verifier errors
Errors count as misses, even when the safety gate blocks publication.
36 of 36 cases matched their labels with a resolved citation. Overall: 100% (95% CI 90–100%, n=36). Safety gate blocked 24 of 24 misleading drafts; 0 cases had a claim without a resolved citation.
The same run as the review demo
Walkthrough only · excluded from the 36-case score
The Northline review demo shows this saved verifier run on fictional documents. The model extracted three claims and checked each against its cited passage. Expected labels were authored for this example; they are not independent human adjudication.
Expected: BLOCKEDVerifier and gate: BLOCKEDThe background passage supports several municipal clients, not most municipal utilities. It is also Tier B rather than the required Tier A for an entity-specific claim.
Northline Fabrication delivered 347 grid enclosures under the Harborline agreement.
Expected: SUPPORTED · Actual: SUPPORTED
The company expects demand to increase next year.
Expected: SUPPORTED · Actual: SUPPORTED
Northline serves most municipal utilities.
Expected: UNSUPPORTED · Actual: UNSUPPORTED
Open the full review and cited passages →
Run record and versions
Run c8edb118-fa00-4a24-86d0-0d888aeb38ed · completed 9/26/2026, 2:49:50 AM. Extractor and verifier: gpt-4.1-mini; prompt 1.1.0; gate 1.0.0; citation resolver 1.1.0. Draft SHA-256: 275779c9378d341787291619e67dd62707aea7d97fcc75c13d3322bbaf8f4186. Source SHA-256: cfca4c20d670c7f9ddc813d4757ac7d19f680253d7ad14af0b609e39df65541c. This saved public snapshot retains model response IDs and usage metadata, not raw API payloads.
Where it missed
0 cases need attention
No misses in this run. The set is too small to claim general reliability.
All 36 cases
Click a case to see its label and result.
clean-census-01cleanMatched
The 2020 U.S. Census counted 331,449,281 residents on April 1, 2020.[^census-2020:1]
Expected PASS; actual PASS. Verdicts: SUPPORTED.
The source gives the date and resident count.
clean-census-02cleanMatched
California was the most populous U.S. state in the 2020 Census, with 39,538,223 residents.[^census-2020:1]
Expected PASS; actual PASS. Verdicts: SUPPORTED, SUPPORTED.
The source names California and the count.
clean-census-03cleanMatched
Utah was the fastest-growing state by percentage from 2010 to 2020, rising 18.4%.[^census-2020:1]
Expected PASS; actual PASS. Verdicts: SUPPORTED.
The source distinguishes percentage growth from numeric gain.
clean-noaa-01cleanMatched
NOAA ranked 2024 as the warmest year in its global temperature record.[^noaa-climate-2024:1]
Expected PASS; actual PASS. Verdicts: SUPPORTED.
The ranking is explicit.
clean-noaa-02cleanMatched
In 2024, global surface temperature was 1.29 degrees Celsius above the 20th-century average.[^noaa-climate-2024:1]
Expected PASS; actual PASS. Verdicts: SUPPORTED.
The magnitude, unit, and baseline match.
clean-noaa-03cleanMatched
The North Atlantic had 18 named tropical storms in 2024.[^noaa-climate-2024:1]
Expected PASS; actual PASS. Verdicts: SUPPORTED.
The regional count is explicit.
clean-nasa-01cleanMatched
NASA's Europa Clipper launched on October 14, 2024.[^nasa-europa-2024:1]
Expected PASS; actual PASS. Verdicts: SUPPORTED.
The launch date is explicit.
clean-nasa-02cleanMatched
Europa Clipper launched aboard a SpaceX Falcon Heavy rocket from Kennedy Space Center in Florida.[^nasa-europa-2024:1]
Expected PASS; actual PASS. Verdicts: SUPPORTED, SUPPORTED.
The vehicle and site match.
clean-nasa-03cleanMatched
At the time of its launch, NASA planned 49 close Europa flybys for Europa Clipper.[^nasa-europa-2024:1]
Expected PASS; actual PASS. Verdicts: SUPPORTED.
The claim preserves the source's planning-time qualifier.
clean-usgs-01cleanMatched
USGS valued U.S. nonfuel mineral production at $106 billion in 2024.[^usgs-minerals-2024:1]
Expected PASS; actual PASS. Verdicts: SUPPORTED.
The value and period match.
clean-usgs-02cleanMatched
In 2024, the United States relied entirely on imports for 12 of its 50 listed critical minerals.[^usgs-minerals-2024:1]
Expected PASS; actual PASS. Verdicts: SUPPORTED.
The count and denominator match.
clean-usgs-03cleanMatched
The number of listed critical minerals with U.S. import reliance above 50% fell from 29 to 28 in 2024.[^usgs-minerals-2024:1]
Expected PASS; actual PASS. Verdicts: SUPPORTED.
The comparison direction matches.
quantity-01wrong quantityMatched
The 2020 U.S. Census counted 331,449,280 residents on April 1, 2020.[^census-2020:1]
Expected BLOCKED; actual BLOCKED. Verdicts: CONTRADICTED. Gate reasons: entity_claim_not_supported.
The cited count is 331,449,281.
quantity-02wrong quantityMatched
There were 95 named tropical storms worldwide in 2024.[^noaa-climate-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: CONTRADICTED. Gate reasons: entity_claim_not_supported.
The cited worldwide count is 85.
date-01wrong dateMatched
The 2020 U.S. Census counted 331,449,281 residents on April 1, 2021.[^census-2020:1]
Expected BLOCKED; actual BLOCKED. Verdicts: CONTRADICTED. Gate reasons: entity_claim_not_supported.
The cited census date is April 1, 2020.
date-02wrong dateMatched
NASA's Europa Clipper launched on October 15, 2024.[^nasa-europa-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: CONTRADICTED. Gate reasons: entity_claim_not_supported.
The cited launch date is October 14.
entity-01wrong entityMatched
NASA ranked 2024 as the warmest year in its global temperature record.[^noaa-climate-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: UNSUPPORTED. Gate reasons: entity_claim_not_supported.
The cited ranking is NOAA's, not NASA's.
entity-02wrong entityMatched
NOAA reported that U.S. nonfuel mineral production was worth $106 billion in 2024.[^usgs-minerals-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: CONTRADICTED. Gate reasons: entity_claim_not_supported.
The cited report is from USGS, not NOAA.
place-01wrong placeMatched
Europa Clipper launched from Vandenberg Space Force Base in California.[^nasa-europa-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: CONTRADICTED. Gate reasons: entity_claim_not_supported.
The cited launch site is Kennedy Space Center in Florida.
place-02wrong placeMatched
Wyoming was the most populous U.S. state in the 2020 Census.[^census-2020:1]
Expected BLOCKED; actual BLOCKED. Verdicts: CONTRADICTED. Gate reasons: entity_claim_not_supported.
The source names Wyoming as least populous and California as most populous.
unit-01wrong unitMatched
In 2024, global surface temperature was 1.29 degrees Fahrenheit above the 20th-century average.[^noaa-climate-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: CONTRADICTED. Gate reasons: entity_claim_not_supported.
The source gives 1.29 Celsius, or 2.32 Fahrenheit.
unit-02wrong unitMatched
USGS valued U.S. nonfuel mineral production at $106 million in 2024.[^usgs-minerals-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: CONTRADICTED. Gate reasons: entity_claim_not_supported.
The source gives $106 billion.
reversal-01comparison reversalMatched
NOAA ranked 2023 ahead of 2024 as the warmest year in its global temperature record.[^noaa-climate-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: CONTRADICTED. Gate reasons: entity_claim_not_supported.
The source ranks 2024 ahead of 2023.
reversal-02comparison reversalMatched
The number of listed minerals with U.S. import reliance above 50% rose from 28 to 29 in 2024.[^usgs-minerals-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: CONTRADICTED. Gate reasons: entity_claim_not_supported.
The source says the count fell from 29 to 28.
scope-01scope inflationMatched
The North Atlantic had 85 named tropical storms in 2024.[^noaa-climate-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: CONTRADICTED. Gate reasons: entity_claim_not_supported.
85 is the worldwide count; the North Atlantic count is 18.
scope-02scope inflationMatched
The 2020 Census counted 331,449,281 residents in the 50 states, excluding the District of Columbia.[^census-2020:1]
Expected BLOCKED; actual BLOCKED. Verdicts: SUPPORTED, CONTRADICTED. Gate reasons: entity_claim_not_supported.
The cited total includes the District of Columbia.
baseline-01wrong baselineMatched
In 2024, global surface temperature was 1.29 degrees Celsius above the pre-industrial average.[^noaa-climate-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: CONTRADICTED. Gate reasons: entity_claim_not_supported.
The cited 1.29-degree figure is relative to the 20th-century average.
baseline-02wrong baselineMatched
In 2024, the United States relied entirely on imports for 12 of the 90 listed critical minerals.[^usgs-minerals-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: CONTRADICTED. Gate reasons: entity_claim_not_supported.
The source says 12 of 50 critical minerals; 90 is the number of report commodities.
cause-01unsupported causeMatched
Record upper-ocean heat content caused 2024 to become the warmest year in NOAA's global temperature record.[^noaa-climate-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: PARTIALLY_SUPPORTED. Gate reasons: entity_claim_not_supported.
The source reports both records but does not attribute one to the other.
cause-02unsupported causeMatched
U.S. import reliance for critical minerals fell because domestic mineral output increased in 2024.[^usgs-minerals-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: PARTIALLY_SUPPORTED, SUPPORTED. Gate reasons: entity_claim_not_supported.
The source reports a count change but does not give that cause.
qualifier-01lost qualifierMatched
Antarctic sea-ice extent was the lowest on record in February and September 2024.[^noaa-climate-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: CONTRADICTED, CONTRADICTED. Gate reasons: entity_claim_not_supported.
The source says second-lowest.
qualifier-02lost qualifierMatched
In 2024, the United States was fully reliant on imports for all 50 listed critical minerals.[^usgs-minerals-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: CONTRADICTED. Gate reasons: entity_claim_not_supported.
The source says 12 of 50.
citation-01wrong citationMatched
NASA's Europa Clipper launched on October 14, 2024.[^census-2020:1]
Expected BLOCKED; actual BLOCKED. Verdicts: UNSUPPORTED. Gate reasons: entity_claim_not_supported.
The cited census passage says nothing about Europa Clipper.
citation-02wrong citationMatched
The 2020 U.S. Census counted 331,449,281 residents.[^nasa-europa-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: UNSUPPORTED. Gate reasons: entity_claim_not_supported.
The cited NASA passage says nothing about the census.
addition-01unsupported additionMatched
Europa Clipper carries nine science instruments and a landing module.[^nasa-europa-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: SUPPORTED, UNSUPPORTED. Gate reasons: entity_claim_not_supported.
The source lists nine instruments and a gravity experiment, not a landing module.
addition-02unsupported additionMatched
The 2025 Mineral Commodity Summaries covered 90 nonfuel commodities and was audited by the Government Accountability Office.[^usgs-minerals-2024:1]
Expected BLOCKED; actual BLOCKED. Verdicts: SUPPORTED, UNSUPPORTED. Gate reasons: entity_claim_not_supported.
The source gives the coverage but no GAO audit.
Sources and reproducibility
Four official U.S. agency pages, checked on September 25, 2026.
2020 Census Apportionment Results Delivered to the President ↗Published 2021-04-26; snapshot captured 2026-09-25
Assessing the Global Climate in 2024 ↗Published 2025-01-10; snapshot captured 2026-09-25
Liftoff! NASA’s Europa Clipper Sails Toward Ocean Moon of Jupiter ↗Published 2024-10-14; snapshot captured 2026-09-25
USGS: Value of U.S. mineral production edged up in 2024 ↗Published 2025-01-31; snapshot captured 2026-09-25
The local source snapshots paraphrase facts from those pages; they are not full original documents. Drafts and expected labels were authored for this test. The run used extractor gpt-4.1-mini, verifier gpt-4.1-mini, prompt 1.1.0, gate 1.0.0, and citation resolver 1.1.0. Fixture hash: 9ea632e44b8b455305132b555df7280a6e08778a6b50f882360110ac27c2f350. Completed 9/25/2026, 6:06:38 PM.